Abstract
We present two fundamentally different approaches to
detect collisions between two point clouds and
compare their performance on multiple datasets. A
collision between points happens if they are closer
to each other than a given threshold radius. One
approach utilizes the main CPU with a k-d tree
datastructure to efficiently carry out fixed range
searches around points in 3D while the other mainly
executes on a GPU using a regular grid decomposition
technique implemented in the CUDA framework. We will
show how massively parallel 3D range searches on a
grid based datastructure on a GPU performs similarly
well as a tree based approach on the CPU with orders
of magnitude less parallelization. We also show how
each method scales with varying input sizes and how
they perform differently well depending on the
spatial structure of the input data.
Users
Please
log in to take part in the discussion (add own reviews or comments).